llama-cpp-canister-release-test

name	llama_cpp_canister-release-test
description	Test a llama_cpp_canister release zip end-to-end
disable-model-invocation	false
user-invocable	true
argument-hint	["release-tag"]
allowed-tools	Bash, Read, AskUserQuestion

Test llama_cpp_canister Release

Tests that a release zip contains everything needed and works end-to-end.

Follow these steps exactly in order. Abort and report on any failure.

1. Download & unzip release

Ask the user which release tag to test. If they say "latest", determine it with:

gh release view --repo onicai/llama_cpp_canister --json tagName --jq '.tagName'

Download and unzip:

rm -rf /tmp/llama_cpp_release_test
mkdir -p /tmp/llama_cpp_release_test
gh release download <TAG> --repo onicai/llama_cpp_canister --dir /tmp/llama_cpp_release_test
mkdir -p /tmp/llama_cpp_release_test/<TAG>
unzip /tmp/llama_cpp_release_test/llama_cpp_canister_<TAG>.zip -d /tmp/llama_cpp_release_test/<TAG>/

All subsequent commands run from /tmp/llama_cpp_release_test/<TAG>/.

2. Follow README setup instructions

Verify dfx

dfx --version

Verify the version is >= 0.31.0. If not, warn the user.

Create conda environment

source /opt/miniconda3/etc/profile.d/conda.sh
conda create -y -n llama_cpp_canister_release_test python=3.11
conda activate llama_cpp_canister_release_test

Install dependencies

cd /tmp/llama_cpp_release_test/<TAG>
pip install -r requirements.txt

3. Deploy

cd /tmp/llama_cpp_release_test/<TAG>
dfx start --clean --background
dfx deploy --network local
dfx ledger fabricate-cycles --canister llama_cpp --t 20

4. Download Qwen2.5 model from HuggingFace

cd /tmp/llama_cpp_release_test/<TAG>
mkdir -p models/Qwen/Qwen2.5-0.5B-Instruct-GGUF
wget -c -O models/Qwen/Qwen2.5-0.5B-Instruct-GGUF/qwen2.5-0.5b-instruct-q8_0.gguf \
  https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q8_0.gguf

Verify SHA256:

echo "ca59ca7f13d0e15a8cfa77bd17e65d24f6844b554a7b6c12e07a5f89ff76844e  models/Qwen/Qwen2.5-0.5B-Instruct-GGUF/qwen2.5-0.5b-instruct-q8_0.gguf" | shasum -a 256 -c

If the checksum does not match, abort immediately.

5. Upload model into canister

cd /tmp/llama_cpp_release_test/<TAG>
python -m scripts.upload --network local --canister llama_cpp \
  --canister-filename models/model.gguf --filetype gguf \
  --hf-sha256 "ca59ca7f13d0e15a8cfa77bd17e65d24f6844b554a7b6c12e07a5f89ff76844e" \
  models/Qwen/Qwen2.5-0.5B-Instruct-GGUF/qwen2.5-0.5b-instruct-q8_0.gguf

6. Run pytest

cd /tmp/llama_cpp_release_test/<TAG>
pytest -vv --network local test/test_qwen2.py

If any tests fail, report the failures but continue to cleanup.

7. Cleanup

cd /tmp/llama_cpp_release_test/<TAG>
dfx stop

Ask the user if they want to remove the test directory and conda environment. If yes:

rm -rf /tmp/llama_cpp_release_test
source /opt/miniconda3/etc/profile.d/conda.sh
conda env remove -y -n llama_cpp_canister_release_test

8. Summary

Report:

• Which release tag was tested
• Whether all tests passed or which tests failed
• Any issues encountered during the test